Messages

Here's the shot. It seems something was going wrong with the deflicker before it started skipping. Several frames had exposure gain inexplicably set 0.05EV-ish higher than those around them, enough to flicker. And these got more frequent towards the end.

PS. If you add

-ext .xmp

to the end of the exifttool command from a couple of posts back, it makes it run much quicker.

I tried this last night on 5D2 using 5th August build (ad21e3450a3c). 5 sec interval, 2 sec max exposure. All other ettr setting default. Shot started at 0.3 secs, 100 iso and all seemed fine. When the exposure time reached 0.6 secs it started dropping the odd .xmp file, about 1 every 10. As the exposure increased more .xmp files got skipped though the frame timing seemed to stay tight. At 1.3 secs it was skipping every other .xmp and after six frames it locked up, battery out to get it restarted.

Guessing that this was due to pushing the timings, I restarted it with a 10 sec interval, all other settings the same, but this time it crashed out after only twenty or so frames. This is odd as I've run with these settings for hours before with no problem at all, so maybe the previous crash had left some mess behind. My previous tests were with an older build, July 21st I think.

I'll give it a try. I'd been erring on the side of caution, not wanting to risk dropped frames.

Point taken.

Another thought. Is there any way the intervelometer could be set to take the next shot as soon as the ettr module has finished processing? Be sure you were shooting as fast as possible then, and the interval could also adapt to longer exposures as light levels dropped.

exiftool -Exposure2012+=-0.5 /path/to/ettr_sequence_folder will knock half a stop off all the exposures.

I was thinking for a shot like this you could set the shutter to between half a second and two seconds. Start around one second at 400 ISO and you've got three stops either way but without letting the shutter get too quick, which would help capture a bit of motion blur, and thus smoother video. But I guess changing ISO that much is undesirable?

Another ETTR example, this time performing admirably in some very changeable conditions.

I've shot with default settings and then used exiftool to scale exposures without loosing the deflicker. Processed through LR+AE. Not tested the dcraw workflow yet, having too much fun shooting. Shots that were next to impossible are now easy!

Is there a way to limit the max shutter speed as well as the minimum? Be great to have a bit more control over shutter angle.

Being able to damp the exposure compensation, so it let the shot get a little darker/lighter as ambient brightness changed, would also be great but I guess quite tricky.

@mvejerslev
I think I know what you mean, these sequences do have a look don't they. They're always going to have a relatively small effective shutter angle, which I'm sure is part of it. But perhaps there's more than that, can't quite put my finger on it, but...

@glubber
After some searching it seems this is simple to do with exiftool. A command of:
exiftool -Exposure2012+=-0.5 /path/to/ettr_sequence_folder
will take 0.5 EV off all the exposure values.

I would be using LRTimelapse for manipulating the xmp files. Adjust some keyframes in LR an LRTimlapse will interpolate beetween them.

I don't know of a way to adjust exposure in LR or LRTimelapse without loosing ML deflicker data, and from tests so far I'd say the ML deflicker works better in this situation than LRTimelapses, not to mention the time saved... Perhaps if we all peck him on his forums, he'll add ML-XMP import to version 3. 
http://forum.lrtimelapse.com/Thread-must-xmp-be-initialized

I've had quite a few tries with ETTR timelapse now and I'm very, very impressed. It's the best ramping solution I've tried, and I've tried a fair few. Most of what I've shot has ended up in the bin due to my crappy photography, or the rebellious British weather, but the two shots bellow have worked out ok. Besides the fly on the lens in the first one, and the motion control going tits up towards the end. But ETTR has performed admirably in all.

These have been shot essentially with default ETTR settings, bar increasing the max exposure time and setting the sidecar to XMP. Processing is limited to static white balance and curves, plus ML deflicker. Ramping over a 13 stop range in the first shot!

I've got a few questions/requests if you don't mind.

Is it possible to process these same files through dcraw, using the deflicker in the xmp files? I'd like to compare results but I'm a bit wet behind the ears when it comes to the command line.

Am I right in thinking that if I tweak the deflicker target down a stop, this will offset the post gain applied by the same value? Would it be possible for this to be a slightly more intuitive scale? Not knowing how the algorithm works setting -4EV to get essentially unity makes no sense to me. And could it be adjustable in finer increments, half or third stops?

A utility that allowed you to offset exposure gain in post, whilst maintaining deflicker, would be very handy. Does anyone know of a way to do this?

I've had a few dropped exposures. Not many but enough to be a concern. Am I pushing it to do a shot every 12 secs with a max exposure of 4 secs on a 5D2? They seem to happen when the exposure is on the long side, but that's an impression, not carefully tested fact.

The resultant RAW files sometimes seem to lack lens correction data, but not always. Easily rectified but I thought worth pointing out.

Is there any way that starting the intervalometer could automatically turn the LCD brightness right down? Be a really handy power saving feature.

My biggest request is the obvious one, speed. The amount of time it takes to process is a limiting factor in these day-to-night type situations. Any little optimisations here would make a big difference.

Once again, I tip my hat to you chaps. And I think Canon ought to send you a very nice christmas card, with a very big cheque inside!

Might it help if the demosaicing algorithm was aware of the actual physical layout of the pixels? We're on the edge of my understanding here but I believe for video the 5D3 uses small clusters of pixels, spaced across the chip. If a standard bicubic interpolation was applied to this, pixels from within the cluster would be weighted equally to those which are actually much further away on the chip. If the debayer algorithm could be weighted to compensate for the specific pixel layout perhaps it would suppress some of the aliasing.

I guess this is pretty obvious, but turning off the buffer display does get me a little extra write speed. With a.d.s 5bad65ff8cb8 build, buffer display off, write speed peaks at 74MB/s and at 1856x1044 25fps I get between 700 and 800 frames. With buffer display on, or using the standard nightly 2013-07-20, write peaks at 73MB/s and at the same resolution I get 600 to 700 frames
Komputerbay 1000x 64GB in a very old and tired 5D2.

@a.d. Cheers for the switch to turn the buffer display off. I couldn't stop watching it. Kept cocking me shots up.

I've always found that third stop jumps need more than just exposure adjustment. It is possible so long as there's no clipping, one big advantage of ETTR is fine control over this, but I have to tweak the contrast too to get a unable shot, and I've not found anything that can automate this.

If we can capture 2144x1076 25 times a sec, and we can dolly this around with digital dolly, then in theory it would be possible to capture the whole raw frame at speeds like 1fps, but with some very odd shutter effects. If you could minimise these and combine it with fine exposure control and ETTR, you'd have a unique tool for timelapse. The 5D3 might be even more impressive!

That would be an incredible feature if it could be done? Wouldn't that also give you finer control over exposure time, and potentially smoother ramping. Not to mention giving the mechanical shutter a rest.

@Euan - Thanks man

@dariSSight - The frame rates in this vary from sync speed down to a shot every 30 secs, and ISOs 100 up to 3200. Horses for courses really. Night scenes were often captured at 3200 ISO, and little or no noise reduction was used in those shots as it created artefacts. But they were captured at least 4K, so noise is partially averaged out at 1080p. And onlining uncompressed 12 bit has kept banding at bay until the very end. The 5D3 was hired in as a spare body for a commission, so I thought I'd give it a go.

Here's a little film I've been working on, using Magic Lantern RAW on 5D2 (and a little 5D3) and a load of timelapse.

Thanks to everyone on here, especially a1ex and a.d. Without MLRAW it would have all been timelapse.

Could I make the case for the return of the old buffer graphic? It may have been a little primitive but it was very easy to see at a glance how much you had left. The new graphic, though actually giving more info, is harder/slower to read and I find it quite distracting during a shot.

In fact, with the 'expected frames' pretty accurate, do we need a buffer graphic at all? If this is just a diagnostic tool for developers that will be disabled on the stable distributions, could we perhaps have an off switch now?

Rec 709 is the colour space of standard HD video. If you're using a 10 bit or better codec and you intend to grade these shots further, you might want to consider using the broadcast limited version [Rec 709 (16-235)]. This way some highlight and shadow detail is retained beyond that visible in AE. However, these files will look flat in apps that don't honour broadcast levels, like QuickTime player and FCP.

I have a 64 GB Komputerbay 1000x that usually gives me a speed of 64-68 MB/s. But on occasion I'll see an almost full buffer just suddenly empty without the write speed registering any real change. Never noticed this behaviour on the Lexar 1000x, but that could just be me not paying attention.

sould i send back and hope for a better one? i ve seen benchmarks up to 110MB/S but on 5D3, is it related?

Sounds like you got a good one there fella. I'd keep hold of it.

You're on a 5D2? I'd try shooting some raw. If you're getting in the upper 60s MB/s then you're doing as well as I am.

Yep, that works, ta.

Finding a few advantages and a few issues doing RAW recording with an Ikan V8000 monitor attached.

When dropping into record the HDMI output doesn't switch to 480p, so the picture stays at it's largest possible size on the monitor. Huge bonus.

Write speed seems slightly slower than with the internal display, but not much, maybe 5%.

The crop rectangle doesn't display properly. I get one of the two scenarios bellow. Actual recordings are not effected though.





EDIT - It seems if the HDMI is plugged in right through the session, from power up to power down, then I get the later. If the screen is plugged in after entering live view, or if it is disconnected and reconnected at any point during the session, then I get the former.

Great as it is to have record on the Set button, being able to select AF-ON was quite useful. It meant you could start/stop record using a standard remote. Really handy when you can only record short clips and need the camera to be perfectly steady from frame 1.

Any chance we could have this back as a menu option?

Yesterdays build (a4ed3721fc44) has a small negative impact on speed for me. I've lost approx 1-2 MB/s compared to 1222c61e8120. Great to have record back on the set button. And no more magenta previews in finder, at least not for me when using RAWMagic b4 on Mac OS 10.8.3.